Portable integrated body surface myoelectricity, stomach / intestine electricity and electroencephalogram examination system
By using a portable integrated system to achieve multi-channel synchronous acquisition and multi-level feature fusion, combined with pulse neural network classification, the problem that a single bioelectric signal is difficult to reflect the synergistic effect of multiple systems is solved, thus improving the accuracy and real-time performance of disease diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-02
- Publication Date
- 2026-03-24
AI Technical Summary
In existing technologies, a single bioelectric signal is insufficient to fully reflect the physiological state of multiple systems working together, resulting in insufficient accuracy and real-time performance in disease diagnosis.
Design a portable integrated system for examining surface electromyography, gastrointestinal electromyography, and electroencephalography. Employ multi-channel synchronous acquisition and preprocessing of signals, combining feature extraction and multi-level feature fusion with a signal processing module, wireless transmission module, and PC terminal, and utilize a deformable self-attention mechanism spiking neural network for classification.
It achieves efficient fusion and accurate identification of multimodal signals, supports real-time and accurate disease detection, and is suitable for dynamic clinical environments.
Smart Images
Figure CN121714286A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of brain-gut axis research and biomedical signal detection, and particularly relates to a portable integrated body surface electromyography, gastric / electrogastrography and electroencephalography diagnosis system. BACKGROUND
[0002] With the continuous development of medical diagnosis technology, the application of single bioelectric signal (such as electrocardiogram, electroencephalogram, and electrogastric) in disease assessment has been relatively mature and widely used in clinical practice. However, many physiological and pathological conditions involve the synergistic action of multiple systems, and it is often difficult to fully reflect the physiological state of the human body by relying on a single signal. For example, the occurrence and development of nervous system diseases, gastrointestinal dysfunction, and mental illness are usually accompanied by the mutual influence of multiple physiological systems. Therefore, the fusion analysis of multiple bioelectric signals is of great significance for in-depth understanding of the pathological mechanisms of these diseases and improving the accuracy of diagnosis.
[0003] In related technologies, signal processing methods (such as Fourier transform and time domain analysis) have limitations in feature extraction and classification, while machine learning and deep learning have shown high accuracy in physiological signal analysis.
[0004] Therefore, there is an urgent need for a diagnosis system that can fuse multiple bioelectric signals and combine advanced signal processing and artificial intelligence algorithms to improve the accuracy and real-time performance of disease detection. SUMMARY
[0005] Therefore, the present application provides a portable integrated body surface electromyography, gastric / electrogastrography and electroencephalography diagnosis system to solve the technical problems existing in related technologies.
[0006] The present application provides a portable integrated body surface electromyography, gastric / electrogastrography and electroencephalography diagnosis system, comprising: A signal acquisition module for multi-channel synchronous acquisition of electromyography signals, gastric / electrogastrography signals and electroencephalography signals, and preprocessing of the electromyography signals, gastric / electrogastrography signals and electroencephalography signals; the preprocessing includes amplification, filtering and digital-to-analog conversion; A signal processing module connected to the signal acquisition module for receiving and integrating the preprocessed electromyography signals, gastric / electrogastrography signals and electroencephalography signals; A wireless transmission module connected to the signal processing module for transmitting the integrated electromyography signals, gastric / electrogastrography signals and electroencephalography signals to a PC end; A PC end for feature extraction and multi-level feature fusion of the integrated electromyography signals, gastric / electrogastrography signals and electroencephalography signals, and inputting the multi-level fused features into a pulse neural network with deformable self-attention mechanism for classification to obtain the physiological change state results before and after acupoint stimulation.
[0007] In one optional implementation, the signal acquisition module includes: The multi-channel electrode includes lead electrodes, reference electrodes, and bias electrodes. Each electrode is connected to multiple body surface sites to simultaneously acquire electromyographic signals, gastrointestinal electrical signals, and electroencephalogram (EEG) signals from the body surface. The signal acquisition unit, connected to the multi-channel electrodes, is used to amplify, filter, and convert electromyographic signals, gastrointestinal electrical signals, and electroencephalogram (EEG) signals into corresponding digital signals.
[0008] In one optional implementation, the signal acquisition unit uses an ADS1299 chip. Each input pin of the ADS1299 chip is connected in series with a protection resistor and a filter capacitor. The fifth pin is connected to the positive terminal of a 3.3V regulated power supply. The first and second pins are directly grounded. The seventeenth pin is connected to the signal processing module through a serial peripheral interface. The sixteenth pin is connected to the wireless transmission module.
[0009] In one alternative implementation, the system further includes: The power supply module is connected to the signal acquisition module, the wireless transmission module, and the signal processing module respectively, and is used to supply power to the signal acquisition module, the wireless transmission module, and the signal processing module.
[0010] In one optional implementation, the wireless transmission module is a BLE105 module, and the twentieth pin of the BLE105 module receives data from the signal processing module through a serial communication interface. The wireless transmission module is also used to transmit the integrated electromyographic signals, gastrointestinal electrical signals and electroencephalogram signals output by the signal processing module to the PC.
[0011] In one optional implementation, the signal processing module uses an STM32F103ZET6 chip. Pins 70 and 102 of the STM32F103ZET6 chip are connected to the signal acquisition unit through a serial peripheral interface, and pin 101 is connected to the wireless transmission module through a serial communication interface.
[0012] In one optional implementation, the PC terminal includes: The feature extraction and fusion unit is used to extract features from the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals to obtain the time-domain features, frequency-domain features, and time-frequency-domain features corresponding to each type of signal; the time-domain features, frequency-domain features, and time-frequency-domain features corresponding to each type of signal are then concatenated to obtain the concatenated fused features; The multimodal feature dynamic selection unit, connected to the feature extraction and fusion unit, is used to perform dimensionality reduction on the spliced fused features using principal component analysis to obtain a dimensionality-reduced feature matrix; self-attention weighting is applied to the feature matrix to obtain a feature matrix highlighting important features; the feature matrix of the highlighting important features corresponding to each type of signal is divided into global features and local features, and weighted fusion is performed according to the weights corresponding to the global features and local features to obtain multi-level fused features; The spiking neural network classification unit, connected to the multimodal feature dynamic selection unit, is used to input the multi-level fused features corresponding to each type of signal into the spiking neural network with a deformable self-attention mechanism for classification, so as to obtain the physiological changes before and after acupoint stimulation.
[0013] In one optional implementation, the PC terminal further includes: The data preprocessing unit is used to perform secondary preprocessing on the integrated electromyography (EMG), gastrointestinal (GE) and electroencephalography (EEG) signals; the secondary preprocessing includes filtering, noise reduction, correction and data synchronization.
[0014] In one optional implementation, the signal processing module further includes: The event detection unit, connected to the wireless transmission module and the spiking neural network classification unit, is used to input the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals into the target detection model to obtain the key time points corresponding to each type of signal. The key time points corresponding to each type of signal are merged to obtain the final set of key time points. The signal segments corresponding to the final set of key time points and the multi-level fused features corresponding to each type of signal are input together into the spiking neural network with a deformable self-attention mechanism for classification.
[0015] In one optional implementation, the PC terminal further includes: The display module is connected to the spiking neural network classification unit and the wireless transmission module, respectively, and is used to display the physiological changes before and after acupoint stimulation output by the spiking neural network classification unit, as well as the integrated electromyographic signals, gastrointestinal electrical signals and electroencephalogram signals output by the wireless transmission module.
[0016] The present invention has the following beneficial effects: (1) The system of the present invention can acquire EEG, EMG and EGG signals simultaneously through multiple channels, provide high-quality bioelectric signals, and ensure the clarity and accuracy of the signals through low-noise amplification technology, making it suitable for precise diagnosis.
[0017] (2) The system of the present invention is designed to be miniaturized and integrated, meeting the needs of portable applications, supporting real-time signal processing and diagnosis, and suitable for rapid and accurate disease detection in dynamic clinical environments.
[0018] (3) The system of the present invention can efficiently fuse signal features from different modalities through a multi-level feature fusion algorithm and a spiking neural network (SNN), thereby improving the recognition accuracy of physiological state changes.
[0019] (4) The present invention combines the target detection Transformer (DETR) model to accurately identify the signal mutation time point caused by the stimulus, optimize the spiking neural network (SNN) classification process, and ensure the accurate processing and classification of key physiological signal segments. Attached Figure Description
[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0021] Figure 1 This is a schematic diagram of a portable integrated surface electromyography, gastrointestinal electrophysiology, and electroencephalography diagnostic system according to an embodiment of the present invention. Figure 2 This is a chip pin diagram of the signal acquisition unit according to an embodiment of the present invention; Figure 3 This is a circuit schematic diagram of a power supply module according to an embodiment of the present invention; Figure 4 This is a chip pin diagram of a wireless transmission module according to an embodiment of the present invention; Figure 5 This is a chip pin diagram of a signal processing module according to an embodiment of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] Figure 1 This invention illustrates a portable integrated surface electromyography, gastrointestinal electrophysiology, and electroencephalography diagnostic system, as shown in the embodiment of the present invention. Figure 1 As shown, it includes: a signal acquisition module, a signal processing module, a wireless transmission module, and a PC terminal.
[0024] Specifically, the signal acquisition module is used for multi-channel synchronous acquisition of electromyography (EMG), gastrointestinal (EGG), and electroencephalography (EEG) signals, as well as preprocessing of these signals. Preprocessing includes amplification, filtering, and digital-to-analog conversion. The signal processing module, connected to the signal acquisition module, receives and integrates the preprocessed EMG, EGG, and EEG signals. The wireless transmission module, also connected to the signal acquisition module, transmits the integrated EMG, EGG, and EEG signals to a PC. The PC performs feature extraction and multi-level feature fusion on the integrated EMG, EGG, and EEG signals, and inputs the fused features into a spiking neural network with a deformable self-attention mechanism for classification, obtaining the physiological changes before and after acupoint stimulation.
[0025] In one alternative implementation, the signal acquisition module includes: a multi-channel electrode and a signal acquisition unit.
[0026] Specifically, the multi-channel electrode includes lead electrodes, reference electrodes, and bias electrodes. Each electrode is connected to multiple body surface sites to simultaneously acquire electromyographic (EMG), gastrointestinal (GE) and electroencephalogram (EEG) signals from the body surface. The signal acquisition unit is connected to the multi-channel electrode to amplify, filter, and perform digital-to-analog conversion on the EMG, GE, and EEG signals to obtain the corresponding digital signals.
[0027] The multi-channel electrodes are used to acquire electroencephalogram (EEG), electromyogram (EMG), and electrogastrogenomic (EGG) signals. Each physiological signal is equipped with an independent reference electrode and a bias electrode to ensure signal stability and anti-interference capability.
[0028] Preferably, the electroencephalogram (EEG) signal uses an 8-lead electrode, the electromyogram (EMG) signal uses a 6-lead electrode, and the gastric EEG signal uses a 4-lead electrode, with each of the three signals equipped with an independent reference electrode and a bias electrode.
[0029] In one optional implementation, to ensure signal accuracy and the integrity of physiological characteristics, the electrode placement positions of different signal sources are designed according to human anatomical characteristics and international standards, specifically as follows: 1) Electrode placement rules for EEG signals: A head-mounted electrode cap is used, and the electrode arrangement follows the international 10-20 system. Specific locations include: Fp1, Fp2 (forehead area) The reference electrode is placed on the earlobe or the midline of the forehead (e.g., Fpz) to provide a reference potential; the bias electrode (DRL, driving right leg electrode) is placed on the earlobe or the hairline area of the forehead to reduce common-mode interference.
[0030] 2) Electrode placement rules for electromyography (EMG) signals: Surface-mount electrodes are used, fixed to the surface of the main muscle groups and placed along the muscle fiber direction, specifically including: Biceps brachii, triceps brachii, quadriceps femoris, gastrocnemius Reference electrodes are placed in bony locations (such as the elbow, knee, or ankle), while bias electrodes are placed in the clavicle or hip to improve signal quality.
[0031] 3) Electrode placement rules for gastric electrical signaling: Abdominal surface electrodes are used to cover the area with the strongest gastric electrical activity. 2-3 cm below the xiphoid process on both sides (corresponding to the antrum of the stomach); 3-5 cm above and below the umbilicus on the midline of the abdomen; The reference electrode is placed in the lower left abdomen near the iliac bone, or in a bony area of the chest where there is no gastrointestinal peristalsis (such as below the clavicle), while the bias electrode is placed in the anterior superior iliac spine or in the abdomen away from the stomach area.
[0032] The system of this invention can ensure the quality of signal acquisition. Through the above-described electrode layout, high-quality acquisition of multimodal physiological signals can be achieved, while avoiding interference from electromyography (EMG) and electrogastric signals on electroencephalography (EEG) signals. The combination of independent reference electrodes and bias electrodes improves the signal's anti-interference capability, ensuring data accuracy and stability.
[0033] In one alternative implementation, such as Figure 2 As shown, the signal acquisition unit uses the ADS1299 chip. Each input pin of the ADS1299 chip has a protective resistor and a filter capacitor connected in series to prevent electrostatic discharge from damaging the chip and to effectively suppress high-frequency interference signals. The fifth pin is connected to the positive terminal of a 3.3V regulated power supply to ensure stable operation with low power consumption. The first and second pins are directly grounded to form a stable analog reference potential. The seventeenth pin is connected to the signal processing module through the UART serial peripheral interface, and the sixteenth pin is connected to the wireless transmission module.
[0034] Among them, the ADS1299 chip is a high-precision analog front-end chip. This chip has multi-channel synchronous acquisition function and supports the amplification and filtering of weak signals such as electromyography (EMG), electrogastrogenomic (EGG) and electroencephalography (EEG) signals.
[0035] In addition, the signal acquisition unit has multi-mode acquisition capabilities, namely: Continuous acquisition mode: Used for long-term recording of EEG, EMG and gastric electrical signals, suitable for sleep monitoring, emotion assessment and analysis of gastric and EMG activity; Triggered Acquisition Mode: Combines external event triggers (such as stimulus signals or patient actions) to acquire signals within a specific time window, suitable for tasks such as depression identification and heart rate variability analysis.
[0036] In one alternative implementation, the system further includes: The power supply module is connected to the signal acquisition module, the wireless transmission module, and the signal processing module respectively, and is used to supply power to the signal acquisition module, the wireless transmission module, and the signal processing module.
[0037] Preferably, the power module includes a lithium battery charging circuit, a battery power detection and protection circuit, and a voltage regulator circuit, such as... Figure 3 As shown.
[0038] The lithium battery charging circuit uses the TP4056 chip as the lithium battery charging management chip. The TP4056 chip has efficient charging management capabilities, automatically controlling the battery charging process to ensure battery charging safety and extend battery life. The specific connection method is as follows: Pins 4 and 8 of the TP4056 are connected to a 5V power supply to provide external power to the charging circuit; pins 6 and 7 are connected in series with an LED and a 1k resistor to indicate the charging status; pin 5 is connected to the positive terminal of the 3.7V lithium battery to charge the lithium battery; pins 1 and 2 are grounded to complete the ground connection of the charging circuit.
[0039] The lithium battery charging process is precisely controlled by the TP4056 chip, ensuring that the battery is charged within a safe voltage range, avoiding overcharging or over-discharging, extending battery life and improving system stability.
[0040] The battery power detection and protection circuit uses a TP4056 chip for management. Its specific connection method is as follows: Pins 1 and 3 of the TP4056 are grounded; pin 2 is grounded via a 4k resistor and is used for battery power detection and protection. This circuit can monitor the battery power in real time to prevent damage caused by over-discharge or over-charge. Simultaneously, the battery protection circuit can flash an LED when the battery power is low to remind the user to charge it in time, preventing the system from malfunctioning due to insufficient battery power.
[0041] The voltage regulator circuit uses the RT9193-3.3V regulator chip to provide a stable 3.3V voltage to the system. The specific connection method is as follows: The RT9193 chip has a 5V power supply connected in series on pin 1 and is isolated by a Schottky diode to ensure stable operation of the voltage regulator chip; pin 1 is connected in series with the positive terminal of a 3.7V lithium battery and a Schottky diode to provide voltage from the lithium battery and output a stable 3.3V power supply after regulation; pin 5 outputs a 3.3V voltage to provide power to the signal acquisition module, signal processing module and Bluetooth transmission module.
[0042] In one alternative implementation, such as Figure 4As shown, the wireless transmission module uses the BLE105 module, which utilizes Bluetooth Low Energy (BLE) technology to achieve wireless transmission of the acquired signals. The twentieth pin of the BLE105 module receives data from the signal processing module via a serial communication interface.
[0043] The wireless transmission module is also used to transmit the integrated electromyographic signals, gastrointestinal electrical signals, and electroencephalogram (EEG) signals output by the signal processing module to the PC.
[0044] In one alternative implementation, such as Figure 5 As shown, the signal processing module uses the STM32F103ZET6 chip. Pins 70 and 102 of the STM32F103ZET6 chip are connected to the signal acquisition module through the serial peripheral interface, and pin 101 is connected to the wireless transmission module through the serial communication interface.
[0045] Preferably, pin 101 (PB14) of the STM32F103ZET6 chip is connected to pin 20 (RXD) of the BLE105 wireless Bluetooth module via a serial communication interface (UART) to achieve real-time data transmission. The STM32F103ZET6 chip communicates with the ADS1299 chip via the serial communication interface (UART) to synchronously acquire and integrate EEG, EMG, and EGG signals.
[0046] The seventeenth pin of the signal acquisition module is connected to the seventieth pin (PA5) and the one hundred and second pin (PB15) of the STM32F103ZET6 chip via UART (Serial Peripheral Interface) for data reading and processing.
[0047] The sixteenth terminal of the signal acquisition module is simultaneously connected to the nineteenth terminal of the BLE105 wireless Bluetooth module to achieve real-time wireless transmission of physiological signals.
[0048] Pin 20 (RXD) of the BLE105 wireless Bluetooth module receives data from the STM32F103ZET6 chip via the UART interface.
[0049] In one alternative implementation, the PC terminal includes: The feature extraction and fusion unit is used to extract features from the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals to obtain the time-domain features, frequency-domain features, and time-frequency-domain features corresponding to each type of signal; the time-domain features, frequency-domain features, and time-frequency-domain features corresponding to each type of signal are then concatenated to obtain the concatenated fused features.
[0050] Specifically, the feature extraction and fusion module performs detailed analysis of electromyography, electrogastric electroencephalography, and electroencephalography signals in the time domain, frequency domain, and time-frequency domain, extracting features that are helpful for signal recognition and classification, and then fusing and aligning these features. The specific steps are as follows: Electromyography (EMG) signal feature extraction: time domain features: mean, variance, peak-to-peak value, root mean square (RMS), etc.; frequency domain features: power spectral density, energy distribution in a specific EMG frequency band (20-450 Hz); time-frequency domain features: short-time Fourier transform (STFT), wavelet packet decomposition (WPT), etc.
[0051] Gastric electrical signal (EGG) feature extraction: time domain features: slow wave period, amplitude, baseline drift; frequency domain features: 3-9 cpm band energy, power spectrum center frequency; time-frequency domain features: continuous wavelet transform (CWT) for gastric motility pattern analysis.
[0052] Electroencephalogram (EEG) signal feature extraction: Time domain features: mean potential, skewness, kurtosis; Frequency domain features: , , , Wave energy; time-frequency domain characteristics: Hilbert-Huang transform (HHT) or multi-scale entropy analysis.
[0053] Feature fusion: First, the time-domain, frequency-domain, and time-frequency-domain features of EEG, EMG, and EGG signals are extracted using the feature extraction module. The feature matrices of each signal are represented as follows: , , These feature matrices are concatenated to obtain a merged feature matrix. : This operation fuses feature data from different modalities into a unified feature matrix, preparing it for subsequent processing.
[0054] The multimodal feature dynamic selection unit, connected to the feature extraction and fusion unit, is used to perform dimensionality reduction on the spliced fused features using principal component analysis to obtain the dimensionality-reduced feature matrix; self-attention weighting is applied to the feature matrix to obtain a feature matrix highlighting important features; the feature matrix of the highlighting important features corresponding to each type of signal is divided into global features and local features, and weighted fusion is performed according to the weights corresponding to the global features and local features to obtain multi-level fused features.
[0055] Specifically, the Multimodal Feature Dynamic Selection (MFDS) unit is used to adaptively select the most effective features for different tasks or samples when processing multimodal data, thereby improving the model's generalization ability and computational efficiency. The specific processing steps are as follows: Principal Component Analysis (PCA) Dimensionality Reduction: In the spliced feature matrix Principal Component Analysis (PCA) is applied to reduce the dimensionality of the data while retaining as much information as possible. PCA uses a transformation matrix... W PCA By projecting high-dimensional data into a low-dimensional space, we obtain the dimensionality-reduced feature matrix Z: Among them, PCA dimensions Determined by the cumulative variance contribution rate: This dimensionality reduction step can reduce computational complexity while preserving the main variation information in the feature data.
[0056] Self-Attention Weighted Assignment: A self-attention mechanism is applied to the dimensionality-reduced feature matrix Z to weight important features. This mechanism determines the importance of features by calculating the query, key, and value matrices, and then weights different features according to the calculated "attention" weights. The calculation formula is as follows: in, Q , K and V These are query, key, and value matrices, respectively. This is the dimension of the key. Through this operation, the model can automatically focus on important features, improving the quality of feature representation.
[0057] Cross-modal Global-Local Feature Fusion (CGFFM): Building upon feature concatenation, PCA dimensionality reduction, and self-attention weighting, the CGFFM module further fuses features from different modalities. This module first divides the features of each modality into global features. and local features Then, based on the relative importance of global and local features, a weighted fusion strategy is used to obtain the final fused features. : in, and These are the weighting coefficients for global and local features, respectively. This fusion strategy effectively combines global and local information based on the characteristics of different modal signals, thereby improving the quality of the fused features and enhancing the system's diagnostic capabilities.
[0058] The spiking neural network classification unit, connected to the multimodal feature dynamic selection unit, is used to input the multi-level fused features corresponding to each type of signal into the spiking neural network with a deformable self-attention mechanism for classification, so as to obtain the physiological changes before and after acupoint stimulation.
[0059] Specifically, the spiking neural network (SNN) classification unit is used to classify multimodal physiological signal features to identify changes in physiological state before and after acupoint stimulation. The specific processing steps are as follows: Time-to-Spike Coding: First, the extracted multimodal physiological signal features are converted into pulse signals using time coding. Time coding is a method of converting continuous signals into discrete pulses, where the time point of each feature is represented by the arrival time of the converted pulse (Spike Time). Assuming the value of a certain feature is... x (t), whose time-coded pulse signal can be represented as: in, t For the time points of change of eigenvalues, The time constant controls the time interval between pulses. Through time encoding, the continuous characteristics of the signal are converted into pulse signals, which are then processed by the pulse neural network.
[0060] Spiking Neural Network (SNN) Processing: After time encoding, the SNN processes the received pulse signals. The basic unit of an SNN is a spiking neuron, which receives pulse signals and reacts according to the time and frequency of the pulses, thereby achieving signal classification. The learning rule of an SNN is based on synaptic time-dependent plasticity (STDP), meaning that the weights between neurons are adjusted according to the time difference of the pulses. The mathematical formula for the STDP rule is: in, The connection weights between neurons, η is the firing time of the neuron, and η is the learning rate. The time difference function is denoted as . The STDP rule enables the SNN to adjust the weights between neurons based on the time difference, thereby classifying multimodal physiological signals. Through this processing flow, the SNN classification module can effectively identify changes in physiological state before and after acupoint stimulation, providing accurate classification results for diagnosis.
[0061] Furthermore, the spiking neural network (SNN) classification module incorporates a deformable self-attention (DSA) mechanism to improve the performance of multimodal physiological signal classification. The specific steps are as follows: Deformable Self-Attention (DSA) Mechanism: Building upon the standard self-attention mechanism, a deformable self-attention (DSA) mechanism is introduced. DSA dynamically adjusts feature weights, making feature selection more flexible and enabling weighted processing for different input features. Specifically, DSA introduces a deformation factor. This is used to control the weight adjustment of different features, thereby achieving dynamic optimization of feature importance. The formula for calculating deformable self-attention is: in, Q , K , V These represent the query, key, and value matrices, respectively. It is a factor that is dynamically adjusted based on the importance of features. Through this mechanism, the model can automatically learn and adjust during training. Optimize feature selection and reduce the impact of redundant features.
[0062] Optimized Feature Selection: The DSA mechanism enables SNNs to focus more on key features and ignore redundant features during classification, thereby improving classification accuracy. By dynamically adjusting feature weights, SNNs can adaptively select the most representative features, enhancing the effectiveness of multimodal physiological signal analysis. By introducing a deformable self-attention mechanism, this module effectively optimizes the feature selection process, improves the classification ability of multimodal signals, and thus provides more accurate analytical results for diagnosis.
[0063] In one alternative implementation, the PC terminal also includes: The data preprocessing unit is used to perform secondary preprocessing on the integrated electromyography (EMG), gastrointestinal (GE) and electroencephalography (EEG) signals. The secondary preprocessing includes filtering, noise reduction, correction, and data synchronization.
[0064] Specifically, the data preprocessing unit performs denoising, filtering, correction, and synchronization processing on the acquired electromyography, electrogastric, and electroencephalography (EEG) signals to ensure signal quality and the accuracy of subsequent analysis. The specific preprocessing steps include: Electromyography (EMG) signal preprocessing: bandpass filtering and notch filtering are used to remove high-frequency noise and power frequency interference; baseline drift correction is performed to remove the slowly varying parts of the signal; the signal is standardized to make its amplitude consistent; abnormal signal detection is performed based on the pathway, and interpolation methods are used to compensate for lost or abnormal signals.
[0065] Electrogastric signal (EGG) preprocessing: bandpass filtering and notch filtering are performed to remove unnecessary frequency components; baseline drift correction and time-domain smoothing are performed to improve signal stability; spectral decomposition is performed to analyze the energy distribution of different frequency bands; abnormal frequency bands and abnormal channels are detected and compensated to ensure signal integrity.
[0066] EEG signal preprocessing: bandpass filtering and notch filtering to remove noise; artifact removal to remove interference caused by artifacts such as eye movement and muscle movement; baseline correction to ensure signal stability; time-frequency decomposition to analyze the time-frequency characteristics of the signal using wavelet transform and other methods; spatial filtering and abnormal signal detection and correction.
[0067] Data synchronization: Time synchronization correction is performed on electromyography, gastric electromyography and electroencephalography signals from different signal sources to ensure alignment of multi-channel signals.
[0068] In one alternative implementation, the PC terminal also includes: The event detection unit, connected to the wireless transmission module and the spiking neural network classification unit, is used to input the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals into the target detection model to obtain the key time points corresponding to each type of signal. The key time points corresponding to each type of signal are merged to obtain the final set of key time points. The signal segments corresponding to the final set of key time points and the multi-level fused features corresponding to each type of signal are input together into the spiking neural network with a deformable self-attention mechanism for classification.
[0069] Specifically, the target detection Transformer (DETR) model identifies the time points of signal abrupt changes caused by stimuli, thus providing a crucial time window for the spiking neural network (SNN). The specific steps are as follows: The event detection module uses the Object Detection Transformer (DETR) model, primarily to identify the time points of signal abrupt changes caused by stimuli. This module directly detects targets on the input signal through a global self-attention mechanism. The DETR model takes time-series signal data as input and outputs the abrupt change position of each signal segment, i.e., the time point of the event. This information is crucial for the subsequent Spiking Neural Network (SNN) classification module, as it provides a critical time window for analysis. The calculation process of the DETR model can be represented by the following formula: in, X The input signal data matrix is used. The Transformer module utilizes a self-attention mechanism to capture temporal dependencies in the signal, thereby identifying abrupt changes. To distinguish between the order and position of the signal, the model also introduces positional encoding.
[0070] The event detection module can efficiently identify abrupt changes in physiological signals, providing accurate time references for subsequent classification and identification processes, thereby improving the diagnostic accuracy of the system.
[0071] The Object Detection Transformer (DETR) model optimizes the Spiking Neural Network (SNN) classification by detecting key time points in EEG, EGG, and EMG signals, ensuring that only the most relevant physiological signal segments are used for final classification. The specific steps are as follows: Key Time Point Selection: The target detection model uses its self-attention mechanism to select abrupt change time points from EEG, EGG, and EMG signals. These time points represent key changes in the signals, providing important information for subsequent analysis and classification. The selected set of time points is denoted as: in, A collection of key time points, It indicates the specific moment of the signal change.
[0072] Optimizing the Spike Neural Network (SNN) classification: Within the detected key time points, the target detection model extracts the corresponding signal segments and provides them to the SNN classification module. The signal segments within this time window are represented as follows: in, For time points The corresponding signal value, This refers to the signal sequence within this time window. By using only signal segments within key time points, SNNs can focus more on signal intervals with important information, thereby improving classification accuracy.
[0073] Reduce computational resource consumption and false positives: By selecting a time window that includes key time points, the processing of irrelevant signals is effectively reduced, thus lowering computational resource consumption and the possibility of false positives. The optimization process is achieved through the following time window selection formula: in, For the optimal time window, The mutation time point The correlation function represents the importance of that time point to the classification task.
[0074] The event detection module independently detects abrupt change time points in EEG, EGG, and EMG signals, and merges the results into a final set of key time points. in, This is the final set of key time points for all modalities. , ,and These are the key time points for EEG, EGG, and EMG signals, respectively. By merging key information from multiple modalities, the DETR module can achieve more comprehensive event detection and improve the system's classification performance in multimodal signals.
[0075] In one alternative implementation, the PC terminal also includes: The display module is connected to the spiking neural network classification unit and the wireless transmission module, respectively, and is used to display the physiological changes before and after acupoint stimulation output by the spiking neural network classification unit, as well as the preprocessed electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals output by the wireless transmission module.
[0076] Specifically, the host computer display module is responsible for receiving and displaying the data transmitted from the wireless transmission module in real time. This module provides an intuitive graphical interface that displays various signal data, analysis results, and diagnostic results. Its specific functions are as follows: Real-time data monitoring: Connects to a wireless transmission module to receive and display electromyography (EMG), gastric electroencephalography (GEG), and electroencephalography (EEG) signal data in real time, ensuring that users can observe signal changes in real time.
[0077] Signal visualization: The collected signal data is visualized through a graphical interface, making it easier for users to understand the signal characteristics.
[0078] Diagnostic results display: Based on the analysis results of the machine learning model, the diagnostic results are displayed in real time, including a comprehensive assessment of physiological signals and the classification or identification results of various physiological states, to help doctors make diagnostic decisions.
[0079] The present invention has the following beneficial effects: (1) The system of the present invention can acquire EEG, EMG and EGG signals simultaneously through multiple channels, provide high-quality bioelectric signals, and ensure the clarity and accuracy of the signals through low-noise amplification technology, making it suitable for precise diagnosis.
[0080] (2) The system of the present invention is designed to be miniaturized and integrated, meeting the needs of portable applications, supporting real-time signal processing and diagnosis, and suitable for rapid and accurate disease detection in dynamic clinical environments.
[0081] (3) The system of the present invention can efficiently fuse signal features from different modalities through a multi-level feature fusion algorithm and a spiking neural network (SNN), thereby improving the recognition accuracy of physiological state changes.
[0082] (4) The present invention combines the target detection Transformer (DETR) model to accurately identify the signal mutation time point caused by the stimulus, optimize the spiking neural network (SNN) classification process, and ensure the accurate processing and classification of key physiological signal segments.
[0083] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A portable integrated system for detecting electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG), characterized in that: include: The signal acquisition module is used for multi-channel synchronous acquisition of electromyography (EMG), gastrointestinal (GE) and electroencephalography (EEG) signals, as well as preprocessing of EMG, GE, and EEG signals. The preprocessing includes: amplification, filtering, and digital-to-analog conversion; The signal processing module, connected to the signal acquisition module, is used to receive and integrate preprocessed electromyographic signals, gastrointestinal electrical signals, and electroencephalogram (EEG) signals. The wireless transmission module, connected to the signal processing module, is used to transmit the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals to the PC. On the PC, features are extracted and multi-level features are fused from the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals. The multi-level fused features are then input into a spiking neural network with a deformable self-attention mechanism for classification, resulting in physiological changes before and after acupoint stimulation.
2. The system according to claim 1, characterized in that, The signal acquisition module includes: The multi-channel electrode includes lead electrodes, reference electrodes, and bias electrodes. Each electrode is connected to multiple body surface sites to simultaneously acquire electromyographic signals, gastrointestinal electrical signals, and electroencephalogram (EEG) signals from the body surface. The signal acquisition unit, connected to the multi-channel electrodes, is used to amplify, filter, and convert electromyographic signals, gastrointestinal electrical signals, and electroencephalogram (EEG) signals into corresponding digital signals.
3. The system according to claim 2, characterized in that, The signal acquisition unit uses an ADS1299 chip. Each input pin of the ADS1299 chip is connected in series with a protection resistor and a filter capacitor. The fifth pin is connected to the positive terminal of a 3.3V regulated power supply. The first and second pins are directly grounded. The seventeenth pin is connected to the signal processing module through a serial peripheral interface. The sixteenth pin is connected to the wireless transmission module.
4. The system according to claim 1, characterized in that, The system also includes: The power supply module is connected to the signal acquisition module, the wireless transmission module, and the signal processing module respectively, and is used to supply power to the signal acquisition module, the wireless transmission module, and the signal processing module.
5. The system according to claim 1, characterized in that, The wireless transmission module uses a BLE105 module, and the twentieth pin of the BLE105 module receives data from the signal processing module through a serial communication interface. The wireless transmission module is also used to transmit the integrated electromyographic signals, gastrointestinal electrical signals and electroencephalogram signals output by the signal processing module to the PC.
6. The system according to claim 1, characterized in that, The signal processing module uses an STM32F103ZET6 chip. Pins 70 and 102 of the STM32F103ZET6 chip are connected to the signal acquisition unit through a serial peripheral interface, and pin 101 is connected to the wireless transmission module through a serial communication interface.
7. The system according to claim 1, characterized in that, The PC terminal includes: The feature extraction and fusion unit is used to extract features from the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals to obtain the time-domain features, frequency-domain features, and time-frequency-domain features corresponding to each type of signal; the time-domain features, frequency-domain features, and time-frequency-domain features corresponding to each type of signal are then concatenated to obtain the concatenated fused features; The multimodal feature dynamic selection unit, connected to the feature extraction and fusion unit, is used to perform dimensionality reduction on the spliced fused features using principal component analysis to obtain a dimensionality-reduced feature matrix; self-attention weighting is applied to the feature matrix to obtain a feature matrix highlighting important features; the feature matrix of the highlighting important features corresponding to each type of signal is divided into global features and local features, and weighted fusion is performed according to the weights corresponding to the global features and local features to obtain multi-level fused features; The spiking neural network classification unit, connected to the multimodal feature dynamic selection unit, is used to input the multi-level fused features corresponding to each type of signal into the spiking neural network with a deformable self-attention mechanism for classification, so as to obtain the physiological changes before and after acupoint stimulation.
8. The system according to claim 7, characterized in that, The PC terminal also includes: The data preprocessing unit is used to perform secondary preprocessing on the integrated electromyography (EMG), gastrointestinal (GE) and electroencephalography (EEG) signals; the secondary preprocessing includes filtering, noise reduction, correction and data synchronization.
9. The system according to claim 7, characterized in that, The PC terminal also includes: The event detection unit, connected to the wireless transmission module and the spiking neural network classification unit, is used to input the integrated electromyography (EMG), gastrointestinal EEG, and electroencephalography (EEG) signals into the target detection model to obtain the key time points corresponding to each type of signal. The key time points corresponding to each type of signal are merged to obtain the final set of key time points. The signal segments corresponding to the final set of key time points and the multi-level fused features corresponding to each type of signal are input together into the spiking neural network with a deformable self-attention mechanism for classification.
10. The system according to claim 5, characterized in that, The PC terminal also includes: The display module is connected to the spiking neural network classification unit and the wireless transmission module, respectively, and is used to display the physiological changes before and after acupoint stimulation output by the spiking neural network classification unit, as well as the integrated electromyographic signals, gastrointestinal electrical signals and electroencephalogram signals output by the wireless transmission module.